The methylation status of O6-methylguanine-DNA-methyltransferase (MGMT) is a crucial predictor of the responsiveness of glioma patients to alkylating agents used in chemotherapy. Current invasive biopsy methods to test MGMT status are expensive and time-consuming. To this end, we propose a deep learning method to assess MGMT status using advanced magnetic resonance imaging (MRI) scans and Isocitrate Dehydrogenases (IDH) status. Specifically, we employed a bifurcated dual-scale convolutional network to capture multi-scale features. To fuse dual modalities by spatial pixel alignment, we redesigned the symphony fusion module, which employs spatial-aware, multi-head spatial attention, and gating mechanisms. In addition, we redesigned a 3D convolutional block attention module to focus on key information in 3D MRI scans. We achieved the best Area Under the Curve (AUC) of 0.6570 using Diffusion Tensor Imaging (DTI-FA) in the UCSF-PDGM dataset. This approach demonstrates that integrating IDH status with DTI-FA significantly enhances MGMT prediction accuracy.